Using @dusty-nv's Jetson inference (https://github.com/dusty-nv/jetson-inference)
Jetson Developer Kit (2gb kit)
Type C power (5V) supply
Ethernet cable
HDMI Cable
Monitor with HDMI cable
Camera (Logitech C270 HD WEBCAM)
Keyboard & Mouse (wireless)
Memory card (more than 32 GB)
Optional: cooling fan, micro-USB cable(for headless mode)
Jetson-Inference With Docker File: https://github.com/dusty-nv/jetson-inference
Datasets:
CT-Scan:-https://www.kaggle.com/mohamedhanyyy/chest-ctscan-images/download
X-ray:-https://www.kaggle.com/jtiptj/chest-xray-pneumoniacovid19tuberculosis/download
1] Gathering all the Accessories:-
We gathered all the Accessories mentioned above from the local market and collected 2GB Developer kit.
2] Preparing for Setup:-
-connect SD card to PC/Laptop
-Download SD card Image (For 2 GB kit)
-Download SD Card Formatter & Install it.
-Quick format the SD card
-Download , Install & Launch ETCHER
-Select Downloaded image, select target device as Memory card then flash ( takes more than 10 min)
3] Setting up Kit:-
Insert the SD card in kit
Attach the Accessories in slots of kit as shown below
- Turn On power supply & wait for system to boot
- When you boot the first time, the developer kit will take you through some initial setup, including:
- Review and accept NVIDIA Jetson software EULA
- Select system language, keyboard layout, and time zone
- Create username, password, and computer name
- Optionally configure wireless networking
- Select APP partition size. It is recommended to use the max size suggested
- Create a swap file. It is recommended to create a swap file
- Review and accept NVIDIA Jetson software EULA
3] Welcome screen:-
4] Downloading Jetson Inference with Docker Container:-
Open Terminal and type following command
git clone --recursive https://github.com/dusty-nv/jetson-inference
Wait until it downloads the container( May take 10-15 mins on slow connection)
Change Directory to jetson-inference using below command
cd jetson-inference
Run the Docker container command, It will ask for the password of your kit. Enter password(in Linux the password you type is not shown) and press enter
docker/run.sh
It may take time in First time running the docker command and ask for models that you want to download. Download default models and presss ok
5] Now download the trained dataset from kaggel, we used the following datasets:-
CT-Scan:-https://www.kaggle.com/mohamedhanyyy/chest-ctscan-images/download
X-ray:-https://www.kaggle.com/jtiptj/chest-xray-pneumoniacovid19tuberculosis/download
-
CT-Scan:-
It has CT-scan Images of lung of Cancer Patients and Normal People
-
X-ray:-
This dataset consist of lung x-ray images of Covid-19 patient, Pneumonia patient, Tuberculosis patient and Normal people
Now Extract the dataset in location:
Jetson-inference/python/training/classification/data/
With dataset name. In that folder create a label.txt file and name all the things that your kit gonna detect
6] Training our Dataset:-
Enter following command
python3 train.py --model-dir=models/ct-scan --batch-size=4 --workers=1 --epochs=35 data/ct-scan
It will take around 8-10 hours for training and Kit gets hot so don’t touch it.
7] Now export the trained model in onnx file
python3 onnx_export.py --model-dir=models/ct-scan
8] Do the same thing with other dataset
9]Now test the project:-
imagenet --model=models/ct-scan/resnet18.onnx --input_blob=input_0 --output_blob=output_0 --labels=data/ct-scan/labels.txt (input location) (Output location)
For webcam based detection:
imagenet --model=models/ct-scan/resnet18.onnx --input_blob=input_0 --output_blob=output_0 --labels=data/ct-scan/labels.txt /dev/video0
10] you can Train the model once more to increase the accuracy or Go ahead and try it with different models to create awesome project. For any help Contact:- [email protected]
We have enclosed all the project required and processed files in the folder.
Replace this data folder with jetson-inference/python/training/classification/data
&
Models folder with
jetson-inference/python/training/classification/models
1] Gathering all the Accessories:-
We gathered all the Accessories mentioned above from the local market and collected 2GB Developer kit.
2] Preparing for Setup:-
-connect SD card to PC/Laptop
-Download SD card Image (For 2 GB kit)
-Download SD Card Formatter & Install it.
-Quick format the SD card
-Download , Install & Launch ETCHER
-Select Downloaded image, select target device as Memory card then flash ( takes more than 10 min)
3] Setting up Kit:-
Insert the SD card in kit
Attach the Accessories in slots of kit as shown below
- Turn On power supply & wait for system to boot
- When you boot the first time, the developer kit will take you through some initial setup, including:
- Review and accept NVIDIA Jetson software EULA
- Select system language, keyboard layout, and time zone
- Create username, password, and computer name
- Optionally configure wireless networking
- Select APP partition size. It is recommended to use the max size suggested
- Create a swap file. It is recommended to create a swap file
- Review and accept NVIDIA Jetson software EULA
3] Welcome screen:-
4] Downloading Jetson Inference with Docker Container:-
Open Terminal and type following command
git clone --recursive https://github.com/dusty-nv/jetson-inference
Wait until it downloads the container( May take 10-15 mins on slow connection)
Change Directory to jetson-inference using below command
cd jetson-inference
Run the Docker container command, It will ask for the password of your kit. Enter password(in Linux the password you type is not shown) and press enter
docker/run.sh
It may take time in First time running the docker command and ask for models that you want to download. Download default models and presss ok